Page 153 - Designing Sociable Robots
P. 153
breazeal-79017 book March 18, 2002 14:7
134 Chapter 9
with one another. At the highest level, behaviors are organized into competing functional
groups (the primary branches of the hierarchy) where each group is responsible for main-
taining one of the three homeostatic functions (i.e., to be social, to be stimulated by the
environment, and to occasionally rest).
Only one functional group can be active at a time. The influence of the robot’s drives is
strongest at the top level of the hierarchy, biasing which functional group should be active.
This motivates the robot to come into contact with the satiatory stimulus for that drive.
The intensity level of the drive being tended to biases behavior to establish homeostatic
balance. This is described in more detail in section 9.4.
The “emotional” influence on behavior activation is more direct and immediate. As
discussed in chapter 8, each emotional response is mapped to a distinct behavioral response.
Instead of influencing behavior only at the top level of the hierarchy (as is the case with
drives), an active emotion directly activates the coordinating behavioral response. It
accomplishes this by sending sufficient activation energy to its affiliated behavior(s) and
behavior groups such that the desired behavior wins the competition among other behaviors
and becomes active. In this way, an emotion can “hijack” behavior to suit its own purposes.
Each functional group consists of an organized hierarchy of behavior groups. At each
level in the hierarchy, each behavior group represents a competing strategy (a collection
of behaviors) for satisfying the goal of its parent behavior. In turn, each behavior within
a behavior group is viewed as a task-achieving entity whose particular goal contributes to
the strategy of its behavior group. The behavior groups are akin to Tinbergen’s behavioral
centers. They are represented as container nodes in the hierarchy (because they “contain”
the competing behaviors of that group). They are similar in spirit to the behavior groups
of Blumberg’s system, however, whereas Blumberg (1994) uses mutual inhibition between
competing behaviors within a group to determine the winner, the container node compares
the activation levels of its behaviors to determine the winner.
Each behavior group consists of a consummatory behavior and one or more appetitive
behaviors. The goal of a behavior group is to activate the consummatory behavior of that
group. When the consummatory behavior is carried out, the task of that behavior group is
achieved. Each appetitive behavior is designed to bring the robot into a relationship with the
environment so that its associated consummatory behavior is activated. A given appetitive
behavior might require the performance of other more specific tasks. In this case, these more
specific tasks are represented as a child behavior group of the appetitive behavior. Each child
behavior group represents a different strategy for achieving the parent (Blumberg, 1996).
Hence, at the behavioral category level, the functional groups compete to determine
which need is to be met (socializing, playing, or sleeping). At the strategy level, behavior
groups of the winning functional group compete for expression. Finally, on the task level,
the behaviors of the winning behavior group compete for expression. As with Blumberg’s

